Microsoft has unveiled its own AI models for the first time. This move expands its relationship with OpenAI, of which it is a major supporter. Microsoft’s AI unit has introduced its first self-developed models, MAI-Voice-1 and MAI-1-preview.
Microsoft unveils AI models
MAI-Voice-1 can generate a minute’s worth of audio in under a second using a single GPU. This model is already used in features like Copilot Daily. Users can listen to the news of the day and receive podcast-style commentary on topics. Furthermore, the system, accessible to everyone through Copilot Labs, allows for customization of settings such as voice tone and speaking style.
Another model, MAI-1-preview, was trained using approximately 15,000 Nvidia H100 GPUs. Microsoft describes this model as a tool that provides meaningful and useful answers to everyday questions. The company is preparing to integrate this model into Copilot and is also testing it on the AI evaluation platform LMArena.
Microsoft’s Head of Artificial Intelligence, Mustafa Süleyman, previously stated that their models are aimed at consumer experience rather than enterprise use. The company’s advertising and user interaction data facilitates the development of more powerful AI models.
In the future, Microsoft aims to establish an ecosystem of specialized models for different use cases. What are your thoughts on this? Share your thoughts with us in the comments section below.
